Simplify setting up big models

[rjplevin]

1. Instead of having to include all the components "manually", we can have a "components" directory from which all .jl files are included. I'm already doing this in the the core with a func defined in Mimi.jl load_comps(dirname::String="./components").

2. @defcomp adds all components to a global registry. For the (presumably majority of) cases where this registry is exactly the list you want loaded, we can define something like add_all_components(model) which does just that. I've already added a topological sort function based on internal connection dependencies. It probably needs some refinement, but it's close. This avoids having to manually add comps in a right order. (Are there any dependency loops where we would have to indicate dependence on the prior timestep? If not we can postpone solving that.) [See also Think about option to create model from all comps in a module #166]

3. For data loading, we can add a little structure to the data directory so that the paths to files are data/$(comp_name)/$(param_name).csv. Then a common load_all_data(model, my_validation_func) could load and test everything as required.

[davidanthoff]

On 1) I'm not so sure. It will for example break all IDE integration because we can't traverse those kind of includes.

On 2), we can discuss that in #166, but I would put that on the backlog because we do have dependency loops everywhere and solving that properly wouldn't be trivial, and I don't think it is a big pain point or something we promised to do in this first round.

On 3, yes we should definitely make things simpler. I've thought a little bit about this in the past, and I think there are a bunch of extra design dimensions to this. Lets put that one on our list of things we need to figure out in one of our next meetings.

[davidanthoff]

We're closing this because composite components essentially address this and we have a separate issue for the parameter loading story.